Related Experiment Video
Updated: Jan 14, 2026
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.6K
Accuracy and reliability of 3D cephalometric landmark detection with deep learning
Boyan Liu1, Chang Liu2, Yutao Xiong2
1Foshan Stomatological Hospital, School of Medicine, Foshan University, Foshan, 528000, People's Republic of China.
European Journal of Medical Research
|October 22, 2025
Summary
An AI-driven model accurately detects 3D landmarks in oral and maxillofacial scans, improving surgical planning. This automated system enhances landmarking efficiency and precision for specialists in complex cases.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Craniofacial Surgery and Orthodontics
Background:
- Accurate three-dimensional (3D) landmark detection is critical for craniofacial growth assessment and surgical planning (orthodontic, orthognathic, plastic, and trauma procedures).
- Current methods often require manual input, which can be time-consuming and prone to variability.
Purpose of the Study:
- To develop and validate an automated 3D landmarking model for oral and maxillofacial regions.
- To assess the model's accuracy, robustness, and generalizability across different imaging modalities (spiral computed tomography [SCT] and cone-beam computed tomography [CBCT]) and landmark types.
Main Methods:
- An optimized lightweight 3D U-Net network architecture was employed for the automated landmarking model.
- The model was trained and validated on a large dataset of 480 SCT and 240 CBCT cases, with further testing on separate datasets (320 SCT, 150 CBCT).
- Evaluation metrics included Mean Radial Error (MRE) and Success Detection Rate (SDR) at 2-4 mm thresholds, alongside coordinate-wise error analysis.
Main Results:
- The AI model achieved an average MRE consistently below 1.3 mm, even in complex cases (malocclusion, missing teeth, artifacts).
- No significant differences in MRE and SDR were found between internal and external datasets for both SCT and CBCT.
- The model significantly improved landmarking proficiency (15.9% for seniors, 28.9% for juniors) and accelerated interaction time (6-9.5 fold).
Conclusions:
- The AI-driven model provides high-precision 3D localization of oral and maxillofacial landmarks, demonstrating utility in complex scenarios.
- It shows potential as a computer-aided tool for accurate and efficient landmark analysis by specialists.
- Prospective clinical validation is recommended to confirm robustness and generalizability across diverse clinical settings and user experience levels.
